Comprehensive skin microbiome analysis reveals …Comprehensive skin microbiome analysis reveals the uniqueness of human skin and evidence for phylosymbiosis within the class Mammalia
Post on 21-Jun-2020
2 Views
Preview:
Transcript
Comprehensive skin microbiome analysis reveals theuniqueness of human skin and evidence forphylosymbiosis within the class MammaliaAshley A. Rossa, Kirsten M. Müllera, J. Scott Weeseb, and Josh D. Neufelda,1
aDepartment of Biology, University of Waterloo, Waterloo, N2L 3G1 ON, Canada; and bDepartment of Pathobiology, University of Guelph, Guelph, N1G2W1 ON, Canada
Edited by W. Ford Doolittle, Dalhousie University, Halifax, NS, Canada, and approved May 9, 2018 (received for review January 30, 2018)
Skin is the largest organ of the body and represents the primaryphysical barrier between mammals and their external environ-ment, yet the factors that govern skin microbial communitycomposition among mammals are poorly understood. The objec-tive of this research was to generate a skin microbiota baseline formembers of the class Mammalia, testing the effects of host species,geographic location, body region, and biological sex. Skin from theback, torso, and inner thighs of 177 nonhuman mammals wassampled, representing individuals from 38 species and 10 mamma-lian orders. Animals were sampled from farms, zoos, households,and the wild. The DNA extracts from all skin swabs were amplifiedby PCR and sequenced, targeting the V3-V4 regions of bacterial andarchaeal 16S rRNA genes. Previously published skin microbiomedata from 20 human participants, sampled and sequenced usingan identical protocol to the nonhuman mammals, were includedto make this a comprehensive analysis. Human skin microbial com-munities were distinct and significantly less diverse than all othersampled mammalian orders. The factor most strongly associatedwith microbial community data for all samples was whether thehost was a human. Within nonhuman samples, host taxonomic or-der was the most significant factor influencing skin microbiota, fol-lowed by the geographic location of the habitat. By comparing thecongruence between host phylogeny andmicrobial community den-drograms, we observed that Artiodactyla (even-toed ungulates) andPerissodactyla (odd-toed ungulates) had significant congruence,providing evidence of phylosymbiosis between skin microbial com-munities and their hosts.
skin microbiome | phylosymbiosis | 16S rRNA | coevolution | mammals
Skin is the largest organ of the body and represents the pri-mary physical barrier between mammals and their external
environment. Characterization of skin microbiota is essential fordiagnosing skin conditions (1), understanding how an animal co-evolves with its microbiota (2), and elucidating interactions betweenmicrobiota and the host immune system (3). Skin microorganismsalso produce compounds that influence animal physiology, such asintraspecific behavior modifying pheromones (4) and volatile or-ganic compounds that contribute to body odor (5–7).Although many studies have characterized the human skin
microbiome, as reviewed by Grice and Segre (8), far less is knownabout the skin microbiome of nonhuman mammals, particularlyfrom studies that employed high-throughput sequencing tech-niques. Multiple studies have been conducted on both wild andcaptive animals to elucidate the roles that host species, geographiclocation, body region, and captivity status exhibit on the skinmicrobiota (9–14). Previous studies of dogs, cats, and bovines havedemonstrated that healthy skin microbiota differ between healthyand diseased or allergic animals (13–17). Analyzing skin samplesof 63 individuals from five primate species revealed that humanaxillae were associated with distinct microbiota, with lower overallmicrobial diversity (9). The authors suggested that differenceswere due to both human hygiene and host–microbe evolution.Skin biopsies and sloughed skin from free-swimming humpback
whales (Megaptera novaeangliae) from the North Pacific, SouthPacific, and North Atlantic oceans demonstrated that core generawere present despite large geographical distances (11). However,skin microbial communities also exhibited shifts between geo-graphic locations and whale satiation states throughout their mi-gration. A survey of bats determined that the host species,geographic region, and site were significant factors influencing skincommunities (10). Microbial diversity of skin and pouch samplesfrom Tasmanian devils identified a strong influence of geographiclocation and revealed significant differences between wild andcaptive populations (18). Together, these studies indicate that bothphylogeny and habitat can impact skin microbial communities.Studies that focus on a wide range of species are important for
characterizing microbiota–host eco-evolutionary patterns, or“phylosymbiosis.” Phylosymbiosis is the process by which thephylogeny of host species parallels the ecological relatedness ofcorresponding microbial communities in a given anatomical lo-cation (19). Although phylosymbiosis does not necessarily implycoevolution of hosts and their microbiota, coevolution may be onemechanism underlying observations of phylosymbiosis. Using 16SrRNA gene sequencing for high-throughput microbial communityanalysis, phylosymbiosis was observed by comparing insect hostphylogeny with gut microbial community dendrograms (19). In
Significance
Skin forms a critical protective barrier between a mammal andits external environment. Baseline data on the mammalian skinmicrobiome elucidates which microorganisms are found onhealthy skin and provides insight into mammalian evolutionaryhistory. To our knowledge, this study represents the largestexisting mammalian skin microbiome survey. Our findingsdemonstrate that human skin is distinct, not only from otherPrimates, but from all 10 mammalian orders sampled. Identi-fying significant similarities between branching of mammalianphylogenetic trees and relatedness trees for their correspond-ing microbial communities raises the possibility that mammalshave experienced coevolution between skin microbiota andtheir corresponding host species.
Author contributions: A.A.R., K.M.M., J.S.W., and J.D.N. designed research; A.A.R. per-formed research; A.A.R., K.M.M., and J.D.N. analyzed data; and A.A.R., K.M.M., J.S.W.,and J.D.N. wrote the paper.
The authors declare no conflict of interest.
This article is a PNAS Direct Submission.
This open access article is distributed under Creative Commons Attribution-NonCommercial-NoDerivatives License 4.0 (CC BY-NC-ND).
Data deposition: The sequences reported in this paper have been deposited in the Se-quence Read Archive (SRA), https://www.ncbi.nlm.nih.gov/sra (mammalian samples: Bio-Project ID PRJNA385010).1To whom correspondence should be addressed. Email: jneufeld@uwaterloo.ca.
This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1801302115/-/DCSupplemental.
Published online June 5, 2018.
E5786–E5795 | PNAS | vol. 115 | no. 25 www.pnas.org/cgi/doi/10.1073/pnas.1801302115
Dow
nloa
ded
by g
uest
on
June
30,
202
0
contrast, demonstration of coevolution between microbial taxa andhost species would require additional strain-level resolution offeredby alternative gene markers with higher phylogenetic resolution,such as DNA gyrase (20). Using such an approach, coevolution wasidentified within the mammalian gut (20). No study has yet in-vestigated phylosymbiosis of mammalian skin microbiota, whichwould be a first step toward demonstrating evolutionary historiesbetween mammals and their skin microbiota.The objectives of this research were to generate a skin micro-
biome dataset for the class Mammalia and to identify correlationsof skin microbiota with species, geographical location, hygiene,and body region. With respect to species–microbiota relationships,we hypothesized that phylosymbiosis occurs between mammalsand their skin microbiota. We tested this by comparing mamma-lian phylogenies with dendrograms generated from their corre-sponding skin microbial communities. Humans were removedfrom the phylosymbiosis analysis because our data reveal starkdifferences between the skin microbiota of humans and all othermammalian species sampled.
Results and DiscussionComparison of Human and Nonhuman Skin Microbial Communities.Bacterial and archaeal microbial communities of 589 mammalianskin samples were characterized by sequencing the V3-V4 regionsof PCR-amplified 16S rRNA genes (SI Appendix, Table S1). Al-together, a total of 22,728 unique operational taxonomic units(OTUs) were obtained that corresponded to 44 prokaryotic phyla(Dataset S1). An indicator species analysis determined that allhuman samples had relatively high proportions of Staphylococcusepidermidis, Corynebacterium, and Propionibacterium acnes (Table1), which is in agreement with previous studies (8, 21, 22). Incontrast, nonhuman animals (“animals”) were associated with soil-associated bacteria, such as Arthrobacter and Sphingomonas, albeitat lower average abundance than the human indicators. Thisfinding was corroborated by a core OTU analysis (SI Appendix,Fig. S1), in which a core OTU was defined as one that was presentin a minimum of 90% of nonrarefied samples. All mammalianclades shared six core OTUs, including Arthrobacter, Sphingomo-nas, and Agrobacterium. Five mammalian orders were analyzedfurther that contained multiple host species and were not com-
posed of cats, dogs, or humans. Each of the orders, except Peri-ssodactyla, had core OTUs that were not shared with any of theother mammalian orders (SI Appendix, Fig. S1). These core OTUsrepresent microbiota that persist despite different geographicallocations and enclosures.The presence of a large proportion of soil microorganisms (SI
Appendix, Fig. S2) may represent bona fide skin microbial com-munities of these animals, yet likely also reflects frequent contactof the skin with the external environment. Although the samplingprotocol for terrestrial mammals did not include a step to rinse offenvironmental microbiota, as has been adopted for amphibianmicrobiota studies (23), future studies might test alternativesampling protocols to access the mammalian skin microbiota tominimize sampling of allochthonous microorganisms. Residentand transient microbiota on skin has been discussed previously forhuman clinical applications (24, 25). Medically, microorganismsare transient if they can easily be disinfected with antiseptics, orremoved with soap and water (25). The sampling methods used inthis study would detect both the resident and transient microbiota.Human samples possessed a distinct microbial community
from all other nonhuman mammals, except for several domesticpets from the order Carnivora (Fig. 1). In addition, human skinwas significantly less diverse than all other mammalian orders,according to both the number of distinct OTUs (Fig. 2A) andShannon indices (6.54 vs. 3.96, P < 0.001) (Fig. 2B), whichconfirms observations from a previous study investigating themicrobiota of five primate species (9). Other orders in whichmicrobial communities group tightly together include Diproto-dontia (kangaroos), Chiroptera (bats), Rodentia (squirrels), andnonhuman Primates (Fig. 1). A subsequent permutational mul-tivariate ANOVA (PERMANOVA) demonstrated that the fac-tor most strongly associated with community variation for allsamples was whether the host was a human (F1, 587 = 37.8, P <0.001) (Fig. 3). Because humans have undergone recent evolu-tionary divergence from other nonhuman primates, such asorangutans (12–15 Mya) and baboons (21–25 Mya) (26), theseresults suggest that modern human practices, such as spendingthe majority of time indoors, frequent bathing, and wearingclothing, may have impacted the diversity and composition ofmeasured skin microbiota. A portion of the higher diversity in
Table 1. Indicator analysis of human and nonhuman animals
Indicator source Indicator value Animal average Human average Consensus lineage
Animal 0.82 15 1 Arthrobacter0.81 20 2 Sphingomonas0.77 6 1 Microbacteriaceae0.74 9 1 Agrobacterium0.73 12 1 Phycicoccus0.70 8 0 Methylobacterium adhaesivum0.70 7 0 Sphingomonas
Human 0.98 4 220 Corynebacterium0.92 7 273 Propionibacterium acnes0.90 2 75 Corynebacterium0.89 1 34 Finegoldia0.85 3 43 Streptococcus0.85 1 26 Corynebacterium0.84 1 25 Corynebacterium0.81 37 163 Staphylococcus epidermidis0.77 0 7 Anaerococcus0.76 1 30 Corynebacterium0.75 1 14 Peptoniphilus0.71 0 15 Propionibacterium acnes
Indicator OTUs were defined as having an indicator value threshold of ≥0.7 and P ≤ 0.05. Reported averagescorrespond to the number of sequences per sample, rarefied to 1,654 reads per sample total.
Ross et al. PNAS | vol. 115 | no. 25 | E5787
MICRO
BIOLO
GY
PNASPL
US
Dow
nloa
ded
by g
uest
on
June
30,
202
0
nonhuman mammals may be due to an increase in the number oftransient microbiota. However, a study of previously uncontactedAmerindians demonstrated that changes in lifestyle, such asliving outdoors, resulted in higher diversity (27), lending furthersupport to this finding that modern human practices may berapidly changing the skin microbiota.By analyzing all samples together, random forest modeling
identified that human and animal samples could be distinguishedcorrectly 98.5 ± 1.2% of the time. The OTUs that contributedmost to the model include Corynebacterium (2.0%), P. acnes(1.2%), Moraxellaceae (1.2%), and Macrococcus (0.8%). Theseorganisms were all within the top 10 most abundant OTUs in adataset of all samples. A single female human back samplegrouped with the majority of the animal samples because of el-evated abundances of Luteimonas, Planomicrobium, and Plano-coccaceae. The animals that were incorrectly classified were housepets, which had elevated levels of Corynebacterium and P. acnes.The specific house pets that grouped predominately with humanslived exclusively indoors. When all pets were removed from thedataset, humans could be distinguished from animals 99.8% ofthe time, which is 78.2-fold better than expected by chance.
Taxonomic Order and Host Geographic Location Influence theMammalian Skin Microbiota. The effects of mammalian taxon-omy, body region, and location were analyzed to elucidatewhether these factors contributed to the detected skin microbiota.Mammalian order had the strongest association with the observedvariation among animal skin communities (PERMANOVA; F9,502 = 11.3, P < 0.001) (Fig. 3), which was followed by geographic
location (PERMANOVA; F4, 507 = 9.6, P < 0.001) (Fig. 3).Random forest modeling was also conducted on a dataset of onlyanimals to determine how well intrinsic factors (e.g., host taxon-omy) and extrinsic factors (e.g., location) could be classified.Animals could be classified best according to their taxonomicorder. This model was correct at classifying animals into theircorresponding order 87.8 ± 5.0% of the time, which is 5.9-foldbetter than expected by chance. Lower taxonomic orders, such asfamily (86.1 ± 3.9%), genus (84.4 ± 4.7%), and species (83.4 ±6.7%), were progressively classified less accurately. This weakerclassification ability may in part be due to smaller number ofsamples per group for training the model.The ability to classify accurately from specific locations may in
part be due to the soil that is present in a given habitat. Indeed, astudy that analyzed the similarity of skin bacterial communitiesbetween salamanders and their environment noted that certaintaxa were shared between the skin microbiota and the abioticenvironment (28), in part due to contact with forest litter. Aprevious study noted that the host was the most important factorinfluencing the skin microbiota of amphibians, whereas geo-graphic location was the second-most important factor (29),which aligns closely with both the PERMANOVA and randomforest model findings from this study.Other factors that have been demonstrated to influence the
human skin microbiota, such as individuality, biological sex, andbody region, exhibited less of an effect on animals sampled inthis study. Both taxonomic order and geographic location wereclassified more accurately than biological sex (65.2 ± 4.5%),body region (39.9 ± 6.0%), or individual animal (36.6 ± 36.7%).
-0.25
0.00
0.25
0.20.0-0.2-0.4
Order
Carnivora
ChiropteraDiprotodontiaLagomorphaPerissodactyla
PrimatesProboscideaRodentiaXenarthra
Axis 1 (8.7%)
Axi
s 2
(6.1
%)
Artiodactyla
Cat living with dog
DogHuman
Other
Cat
Fig. 1. Ordination (PCoA) generated by using the Bray–Curtis dissimilarity metric for each of the three body locations sampled. Samples are coloredaccording to mammalian order. (Inset) Ordination colored according to human and nonhuman samples.
E5788 | www.pnas.org/cgi/doi/10.1073/pnas.1801302115 Ross et al.
Dow
nloa
ded
by g
uest
on
June
30,
202
0
This inability to classify individuals is in contrast to humanstudies that have shown that individuality is one of the mostimportant factors influencing the human skin microbiota, al-though several of these studies used >15 samples per individual(30–32). It is therefore still possible that the skin microbiota arerelatively unique among individual animals, but this cannot beobserved with only three samples per animal.To address whether biological sex influenced the skin micro-
biota within a species, cat (n = 48), dog (n = 35), and horse (n =68) samples were analyzed because they contained a relativelylarge number of sampled individuals and a similar number ofsamples from both biological sexes. Biological sex was not a sig-nificant factor for any of these species (PERMANOVA; cat:F1, 15 = 1.15, P = 0.20; dog: F1, 11 = 0.79, P = 0.77; horse: F1, 20 =0.94, P = 0.44). All of the domestic cats and dogs were neutered orspayed, so this would not have an influence on the analysis. Al-though the horses had both neutered/spayed and intact individ-uals, whether the horse had been neutered or not did not exhibit asignificant difference on the skin microbiota (PERMANOVA;F1, 20 = 1.09, P = 0.34). The only animal where biological sexexplained the most variation was the red kangaroo (PERMANOVA;F1, 16 = 2.21, P = 0.002), which was also analyzed becausethis species exhibited a visual separation between males and fe-males in an ordination (principle coordinates analysis, PCoA)(SI Appendix, Fig. S3). Larger variations in the microbiotaamong different body regions were observed within Peri-ssodactyla (PERMANOVA; F2, 121 = 4.26, P < 0.001) (Fig. 3) andProboscidea (PERMANOVA; F2, 12 = 2.38, P = 0.02) (Fig. 3) thanwithin other orders. The overall low effect from body region islikely due to the body regions sampled. The back, inner thigh, andtorso are all covered with hair. A previous study on dogs demon-strated that fur-covered regions had higher species richness anddiversity compared with mucosal surfaces (13). Therefore, sam-
pling mucosal surfaces would be expected to result in more dis-tinct differences between body regions within a species.To ensure that the importance of the host’s taxonomic order
on skin microbiota was not overly influenced by orders withfewer samples and locations, the three orders with a large numberof samples—Artiodactyla, Carnivora, and Perissodactyla—wereanalyzed with all other orders removed. Each of these orders hadsamples sourced from six to eight different locations. Removingorders with fewer samples increased the influence of order(PERMANOVA; F2, 385 = 15.1, P < 0.001) and decreased theimpact of geographic location (PERMANOVA; F3, 384 = 8.7,P < 0.001). Therefore, the effect on microbial communitiesexhibited by the mammalian host exists despite varying geographiclocations, and cannot be fully attributed to certain species onlybeing sampled in a single location.
Pets and Humans.Although the majority of animals possessed skinmicrobial communities that were distinct from humans, a subsetof pets grouped with humans in ordination space (Fig. 1). Inparticular, of the 17 animal samples that grouped with humans,15 were from indoor housecats, whereas the remaining twosamples belonged to the backs of dogs that were frequentlybathed and groomed (SI Appendix, Table S2). In total, 75% ofthese samples belonged to animals that were owned by humanswho were also sampled for this study. All cats with similar mi-crobial communities to humans had at least two of the threesampled body locations possess this “human” community com-position (SI Appendix, Table S2), whereas the two dogs only
A
B
Fig. 2. Boxplots of diversity indices for 10 mammalian orders and humans,including both number of OTUs (A) and Shannon indices (B).
Probos
cidea
Chirop
tera
Diproto
donti
aRod
entia
Carniv
ora
Carniv
ora(pe
ts rem
oved
)
Artioda
ctyla
Peris
soda
ctyla
Primate
sAll s
ample
s
Human
s remov
ed
Neutered or spayed
Order
Cage or house
Individual
Location
Genus
Species
Brushing
Family
Bathing
Diet
0 10 20 30
F statistic
Human versus non-human animal
Biological sex
Age range
Body region
= Not applicable
Fig. 3. Heatmap summarizing the significant metadata factors correlatingwith the observed skin microbiota for sampled individuals from mammalianorders. Categories with higher PERMANOVA F statistics have higher varia-tion in community dissimilarity. Gray regions of the heatmap representcategories that do not apply. Samples and categories are clustered accordingto Bray–Curtis dissimilarities.
Ross et al. PNAS | vol. 115 | no. 25 | E5789
MICRO
BIOLO
GY
PNASPL
US
Dow
nloa
ded
by g
uest
on
June
30,
202
0
possessed the human microbial community on their backs. Theremaining 11 dogs had similar communities to the other animals,as did all cats that lived outdoors on farms and a single cat thatlived in the city, without a dog (Fig. 1, Inset). Interestingly, 11 of12 indoor cats that lived with a dog possessed similar microbialcommunities to the other animals. It may be that owning a dogresults in an influx of soil microbiota into the home, which inturn is transferred to indoor cats through contact with the builtenvironment or personal contact. Indeed, previous studies showthat owning a dog shifts the human microbiota as well as mi-croorganisms detected on built environment surfaces (33, 34).Because our study only sampled skin from 13 dogs and 19 cats,future studies might include a larger sample size of animals thatwould help further elucidate how owning a dog impacts the skinmicrobiome of other inhabitants within a household.
Predicted Functions of Skin Microbiota Vary Between Human andAnimal Samples. Predicted functions based on detected prokaryotictaxa were determined using Functional Annotation of ProkaryoticTaxa v.1.0 (FAPROTAX) (35), revealing several conserved func-tions on mammalian skin (Fig. 4). Many of these match functionswere predicted from human samples from the Human MicrobiomeProject consortium that were characterized by metagenomic se-quencing (36, 37). Animal symbionts and human pathogens wereexpected because the samples were derived from mammalianhosts. Urea is a component of sweat and provides a nitrogensource, which could explain ureolysis as a predicted skin function,although not all animals sweat in hair-covered regions (38).There were several functions that were significantly different
between humans and animals. In accordance with lower diversity(Fig. 2), humans had significantly fewer predicted functions (34.2 ±8.4 compared with 51.8 ± 9.4 in animals; P < 0.001). Animals mayalso have a higher number of predicted functions because they had
more soil bacteria, which may have more annotated predictedfunctions than skin bacteria. Humans had elevated levels of pre-dicted manganese oxidation. Human sweat contains 100 ppbmanganese on average, which would result in ∼200–300 mg ofmanganese secreted each day (39). This concentration is lowcompared with other trace metal elements, such as zinc andcopper (40). Manganese oxidation was predicted to occur from thecore human OTU P. acnes (41). In contrast, animals had higherlevels of predicted functions involved in the nitrogen cycle andsingle-carbon compound degradation. Methanol oxidation wasattributed to the core OTUs affiliated with Arthrobacter, andmethylotrophy with Methylobacterium OTUs, according to theFAPROTAX database. Nitrogen respiration was associated withnumerous organisms, such as Paracoccus and Pseudomonas. Un-derstanding if skin microbiota contribute to nitrogen or manga-nese cycling is important because deficiencies in manganese andnitrogenous compounds, such as nitric oxide, have been associatedwith delays in wound healing (42). Metabolites have also beenlinked to aging in mammals due to changes in oxidative stress (43).Predicting functions based on taxonomy is the first step to
elucidating how biochemical processes from skin microbiota areinfluencing host skin health. Many advances have been made ingut microbiome research that have demonstrated microorgan-isms within the gastrointestinal tract can influence digestion (44),provide vitamins and amino acids that the host is unable tosynthesize alone (45), and impact neurological function throughthe gut–brain axis (46). Although it is currently unknown how thebiogeochemical processes of skin microbiota influence their host,determining which processes are occurring on skin is a first steptoward understanding how these microbial functions impactdermatological health. Future studies using metagenomic se-quencing may help confirm which of these predicted conservedmicrobial functions are core to mammalian skin, versus functionsthat are variable among different host species.
Phylosymbiosis Is Evident Within the Orders Perissodactyla andArtiodactyla. Phylosymbiosis would result in closely relatedclades of animals being associated with similar microbial com-munities (19). This effect can be measured using Robinson–Foulds and matching-cluster values, which compare the congru-ence between two relatedness trees. A score of 0 indicates that thetrees are identical, whereas a score of 1 indicates there is nocongruence between the two trees. The matching-cluster methodis statistically more robust than Robinson–Foulds and provides awider distribution of scores between 0 and 1 because it calculatesthe total composition of the subtrees (47). Previous studies dem-onstrate that shifts in microbial communities have matched hostevolution within insects (19, 48), which were more apparent at the99% OTU clustering threshold. In this study, comparisons weremade at both the 97% and 99% thresholds, using Bray–Curtis,unweighted, and weighted UniFrac distance metrics.Comparing host mammalian phylogeny to dendrograms of
corresponding skin microbial communities for each mammalianhost species indicates that relationships among skin microbiotafrom animals of the orders Perissodactyla and Artiodactyla weresimilar to their corresponding host evolutionary relationships (Fig.5). Perissodactyla exhibited phylosymbiosis with all thresholds anddistance measures; the only discrepancy in each test case was theskin microbial communities of domesticated horses and Prze-walski’s horses (Fig. 5A). The split between the equestrian andrhinoceros clades cannot be attributed to differences in location,such as farm or zoo habitats, because the Przewalski’s horses weresourced from the Toronto Zoo. Artiodactyla (Fig. 5B) possessed anormalized matching cluster score of 0.38 (Table 2), and demon-strated significant congruence with the host phylogeny and micro-bial dendrogram with both the Bray–Curtis and weighted UniFracmetrics. The largest discrepancy was noted with the sequences fromthe goats, which grouped with the giraffe and reindeer rather than
Fig. 4. Barplot of predicted functions based on FAPROTAX database. Starsindicate P < 0.05 among mammalian and human samples after Bonferronicorrection was applied. Error bars denote the SD of animal (n = 512) andhuman (n = 77) samples.
E5790 | www.pnas.org/cgi/doi/10.1073/pnas.1801302115 Ross et al.
Dow
nloa
ded
by g
uest
on
June
30,
202
0
the sheep. In addition, the host species did not cluster according tothe geographic locations from where they were sourced. In con-trast, the order Carnivora (Fig. 5C) did not exhibit significantphylosymbiosis and the cat and dog clades did not have distinctmicrobial communities for either metric of relatedness or dis-tance metric. This observation did not change when all cat anddog samples were removed (Table 2). Therefore, the microbialdendrograms were not being unduly influenced by householdanimals that undergo frequent grooming and spend the majority oftime indoors. It is possible that phylosymbiosis may be more stronglyobserved within clades of animals that share similar diets or havesimilar management. All of the sampled animals within Peri-ssodactyla and Artiodactyla were herbivores that graze on localgrasses or hay. In contrast, the animals within the Carnivora orderhad a more diverse diet among member species, such as herbivorousgiant pandas, carnivorous lions, and pets that were fed an omniv-orous diet. Similar to how diet influences the gut microbiota (49), itmay be that the skin microbiota is impacted by diet, as has beenshown for skin microbial communities of amphibians (50).No significant phylosymbiosis was observed when all mamma-
lian orders and humans were analyzed together, except for theunweighted UniFrac measure analyzed using normalized Robinson–Foulds values at the 99% threshold (Table 2). Although, in thissingle case, animals could be matched significantly better than100,000 randomized trees of the 38 species, there was very littlecongruence observed (Table 2), as indicated by the normalizedRobinson–Foulds score of 0.93 and the normalized matchingcluster score of 0.99. When humans were removed from this dataset,congruence increased modestly, although the host phylogeny andbacterial dendrogram still exhibited little congruence.
Although previous studies have been able to demonstratephylosymbiosis, they did so under highly controlled laboratoryconditions and with fecal samples (19, 49, 51). Skin represents amore transient environment that is influenced by shedding andcontact with other surfaces. The animals in this study had severalconfounding factors, such as different locations and age (DatasetS2). In our study, it is possible that more distinctive congruencewould be observed if mammals were sampled at similar time pointsin their life history and inhabited the same geographic location.Additionally, the potentially transient soil microorganisms thatwere more abundant on nonhuman mammalian skin may maskphylosymbiosis when all sequenced OTUs are being considered asa community in the phylosymbiosis analysis (SI Appendix, Fig. S1).Future studies should potentially sample before and after washingthe skin to observe how this treatment would influence the analysis.We postulate that reducing the number of transient auxiliary or-ganisms from the environment would strengthen the finding ofphylosymbiosis because allochthonous microorganisms that do notcoevolve with a host would be removed from the analysis.
Archaea Are Present on Mammalian Skin. Including all human data, ar-chaea comprised only 6,509 of all 6,550,625 nonrarefied sequencesanalyzed in this study (0.1%) (Dataset S1). Several archaeal cladeswere represented, including salt-tolerant Halobacteria, the metha-nogen Methanobrevibacter, and the ammonia-oxidizers of the Thau-marchaeota (Dataset S1). Methanogens likely represent fecalcontamination because Methanobrevibacter spp. are the dominantarchaea present in the gut (52). However, Halobacteria and thau-marchaeotes, such as Nitrososphaera, have the potential to be resi-dent skin microbiota. Halobacteria are able to tolerate the salt
p = 0.002p = 0.009nMC = 0.33 nRF = 0.33
Host phylogeny
Microbiotadendrogram
0.5
White rhinoceros
Przewalski's horse
Pony Pony
Donkey Donkey
Indian rhinoceros Indian rhinoceros
Horse
HorsePrzewalski's horse
White rhinoceros
0.5
Spotted hyena
African lion
Cat
Cheetah
Dog
Arctic wolf
White lion
River otter
nMC = 0.79p = 0.27
nRF = 0.83 p = 0.30
Host phylogeny
Microbiotadendrogram
White lion
African lion
Cat
Cheetah
Dog
Arctic wolf
River otter
Giant pandaGiant panda
Spotted hyena
0.50.5
Sheep
Aoudad sheep
Bovine
Reindeer
Bactrian camel
Rothschild giraffe Rothschild giraffe
A Perissodactyla C CarnivoraB Artiodactyla nMC = 0.38p = 0.001
nRF = 0.71p = 0.03
Bactrian camel
Goat
Sheep
Sable antelope
Sable antelope
Aoudad sheep
Reindeer
Cape elandCape eland
Goat
Bovine
Host phylogeny
Microbiotadendrogram
Alpaca Alpaca
0.5 0.5
Fig. 5. Dendrograms for species-specific skin microbiota generated using the Bray–Curtis distance metric and a 99% OTU threshold, compared with theknown host phylogenies of Perissodactyla (A), Artiodactyla (B), and Carnivora (C). Congruence was measured using normalized Robinson–Foulds (nRF) scoresand normalized matching cluster (nMC) scores. Bootstrap values for all microbiota dendrogram branch points were ≥98% and are not shown. Mammal,bovine, goat, and sheep images courtesy of Tracey Saxby, John C. Fisher, Jane Hawkey, and Tim Carruthers, Integration and Application Network, Universityof Maryland Center for Environmental Science (ian.umces.edu/imagelibrary/). Alpaca image courtesy of Meaghan Mechler (artist).
Ross et al. PNAS | vol. 115 | no. 25 | E5791
MICRO
BIOLO
GY
PNASPL
US
Dow
nloa
ded
by g
uest
on
June
30,
202
0
concentrations from sweat (53). Putative ammonia-oxidizingarchaea have also been reported on human skin (54, 55). Eventhough few archaeal sequences were present in our data overall,archaeal sequence proportions, normalized by the number of in-dividuals sampled, were unevenly distributed among sampledanimal species (SI Appendix, Fig. S4). For example, averaged ar-chaeal sequence proportions were highest for cape elands (26.1%of all summed archaeal species averages), olive baboons (12.9%),the sable antelope (10.9%), and bovine (5.8%). The methanogensfrom the phylum Euryarchaeota were the dominant archaeal clade,which was expected because the animals with the most archaealsequences were predominately ruminants. Averaged thaumarch-aeotal sequences were highest for groundhogs (0.9% of all sum-med archaeal species averages), the swamp wallaby (0.6%), olivebaboons (0.5%), and the pony (0.5%). Together, these resultsexpand the known range of putative ammonia-oxidizing archaeaon skin beyond human beings.Our study likely underestimated archaeal abundance and
diversity because of primer mismatches to archaeal 16S rRNAgenes (56). Indeed, the performance of the Pro341F/Pro805Rprimer set was analyzed using TestPrime v1.0 on the SILVAdatabase (SI Appendix, Table S3). Only 64.8% of archaeal 16SrRNA genes had zero mismatches to the primer set used in thisstudy. Only 11.9% of ammonia-oxidizing thaumarcheota ref-erence sequences had zero primer mismatches, in contrast to85.7% of all bacterial sequences. When the number of mis-matches was increased to two, 94.9% of all archaea and 95.5%of thaumarcheotes were targeted. A recent study on the gutmicrobiota of great apes that used both universal prokaryoticand archaea-specific primers determined that the distribution,diversity, and prevalence of archaea in mammalian gut samplesis underestimated by up to 90% (56).
Limitations. The majority of the animals were collected based onopportunistic availability. For example, animals from the TorontoZoo were sampled during routine veterinary check-ups. The natureof sample collections resulted in an inability to collect from anequal number of representatives from each host taxonomic orderand species. For example, the following species only had a single
representative animal sampled: alpaca, beaver, pony, sheep, sa-ble antelope, spotted hyena, swamp wallaby, and two-toed sloth.Although we recognize that no significant conclusions can bemade about a single host animal within a species, these animalswere included in the analysis to maximize coverage of eachmammalian order. Much work remains to be conducted withineach species to determine intraspecific effects of individuality,body region, and biological sex.Animal skin microbiota were sampled throughout the calendar
year and all samples were frozen until DNA extraction. It ispossible that the skin microbiota of outdoor animals may un-dergo seasonal shifts, especially between the relatively coldwinter and warm humid summer in Canada; however, this cannotbe tested using a single sampling time for each animal. Futureinvestigations should sample the same individuals temporally todetermine if changes in temperature and resulting skin secretionlevels influenced microbial community composition and di-versity. Moreover, the significant difference in geographic lo-cation that was observed may be more pronounced if animalswith greater geographic distance were sampled. All of the an-imals were sampled in Southern Ontario. Sampling the samespecies from multiple continents is hypothesized to result inmore pronounced variations in communities according to lo-cation due to significant changes in extrinsic factors, such as soilmicroorganisms.A limitation of any amplicon study is biases that arise from
primer selection. This study targeted the V3-V4 16S rRNA re-gion. The relative abundances of common skin organisms, suchas Propionibacterium and Staphylococcus, may differ in studiesthat select another portion of the 16S rRNA gene, such as theV1-V3 region (57). Although these biases have been discussed indetail elsewhere (57–60), the V3-V4 region can accurately rep-resent skin microbial communities (61) and is therefore expectedto have influenced the results only minimally.The rodents collected in this study were sourced from both
wild and deceased animals. Although samples from deceasedanimals still grouped with the remaining live animals (Fig. 1),high skin microbial diversity (Fig. 2) may be in part related toinitial changes in skin community from decomposition (62).
Table 2. Phylosymbiosis analysis of selected mammalian clades
Robinson–Foulds Matching cluster
Clade Distance metric 97% threshold 99% threshold 97% threshold 99% threshold
All samples BC 0.97 0.97 1.00 0.98UU 1.00 0.93* 1.00 0.99WU 0.97 0.97 1.00 1.00
All mammals: humans removed BC 0.93* 0.94* 0.98 0.91UU 0.96* 0.94* 0.96 0.98WU 0.93* 0.93* 1.00 1.00
Artiodactyla BC 0.71* 0.71* 0.38* 0.38*UU 0.86 0.86 0.70 0.40WU 0.71* 0.71* 0.48* 0.43*
Carnivora BC 0.83 0.83 0.61 0.79UU 1.00 1.00 0.79 0.79WU 0.83 0.83 0.57 0.57
Carnivora: pets removed BC 1.00 0.75 0.78 0.61UU 1.00 1.00 0.78 0.78WU 0.75 0.75 0.50 0.50
Perissodactyla BC 0.33* 0.33* 0.33* 0.33*UU 0.33* 0.33* 0.33* 0.33*WU 0.33* 0.33* 0.33* 0.33*
The normalized Robinson–Foulds scores and normalized matching cluster scores were calculated at the 97% and 99% clusteringthresholds based on the Bray–Curtis distance metric (BC), the unweighted UniFrac distance metric (UU), and the weighted UniFracdistance metric (UW).*Significant (P ≤ 0.05) scores.
E5792 | www.pnas.org/cgi/doi/10.1073/pnas.1801302115 Ross et al.
Dow
nloa
ded
by g
uest
on
June
30,
202
0
Nonetheless, deceased rodents were collected within 24 h ofdeath and did not have any visible injuries that would result ininternal microorganisms from the gastrointestinal tract contam-inating the skin.
ConclusionOur findings demonstrate that human skin is distinct, not onlyfrom other primates, but from all 10 mammalian orders sampled.Human samples were dominated by S. epidermidis, Corynebacterium,and P. acnes. Given the recent evolutionary divergence of humans asdistinct species from other nonhuman primates, these results suggestthat modern human practices, such as living within a built environ-ment, wearing clothing, and washing with soap, have strongly im-pacted the diversity and composition of the skin microbiota that canbe sampled with sterile swabs. Microbial communities from non-human mammals were significantly more diverse and were associ-ated with higher levels of traditional soil-associated OTUs that likelyrepresent a combination of resident skin microbiota and transientorganisms collected temporarily from their enclosure or naturalhabitat. Despite these environmental influences, phylosymbiosiswas detected in two herbivorous orders: Perissodactyla andArtiodactyla. These results provide first evidence that skin micro-bial communities assemble in a species-specific manner on mamma-lian hosts, which may be initial evidence that coevolution is occurringbetween mammalian hosts and skin microbiota, presumably throughmaternal inheritance.
Materials and MethodsEthics. This study was approved by the Office of Research Ethics (ORE) at theUniversity of Waterloo (A-15-06). Detailed ethical consent for the humanportion of this study was reported previously under ORE no. 20993 (63). Thefollowing minimally invasive procedures were conducted in accordance withthe approved documentation and no animals were harmed throughoutthis study.
Sample Collection. Species from 10 orders of the class Mammalia weresampled to characterize the distribution of microorganisms on skin (SI Ap-pendix, Table S1). Both males and females were included for each species,when available, to account for variations in hormone levels and secretionsthat are known to affect microbial communities (21). A spectrum of habitatsand hygiene practices were also included, ranging from frequent groomingof pets and farm animals, to animals in captivity and the wild. Completeinformation on the biological sex, age, diet, location, health history, grooming,and antibiotic administration were collected (Dataset S2). The inclusion of ani-mals that had received antibiotics in the previous 6 mo was minimal (32 animals;18%), of which 10 received antibiotics in the 2 mo preceding sampling (5.6%).The results indicate that this exposure did not influence measured microbial di-versity of those individuals within each species. Additionally, animals that hadreceived antibiotics did not group separately within ordination space from thosethat had no history of antibiotics. Therefore, these animals were included tostrengthen analyses, such as phylosymbiosis and PERMANOVA. Animals thatreceived antibiotics include the alpaca, aoudad sheep, bovine, olive baboons,cats, dogs, horses, and the Indian rhinoceros.
Animals were sampled from multiple locations in Southern Ontario fromNovember 2015 to September 2016: The African Lion Safari, Kitchener–Waterloo Humane Society (KWHS), Toronto Zoo, pet owners, and from localfarms sourced from the University of Guelph. Animals from the Toronto Zooand African Lion Safari were sampled when they were brought in for regularhusbandry practices. Additional companion animals were obtained fromvolunteers who were recruited by word of mouth. The KWHS supplied wildanimals that were collected by KWHS staff within 24 h of death; the speci-mens were stored in plastic bags in a −20 °C freezer until sampled.
The back, torso, and inner thigh regions of 177 nonhuman mammals werecollected using sterile foam swabs (Puritan) according to a previously pub-lished protocol (63). In addition, we included data from 77 equivalentsamples (the right and left inner thigh were sampled from each participant)from 20 human participants from a previous study that were sampled fromNovember 2015–February 2016 (63) in the analysis for comparison purposes[Sequence Read Archive BioProject ID PRJNA345497 (https://www.ncbi.nlm.nih.gov/sra/PRJNA345497)], for a total of 589 samples. These regions werechosen to capture both moist and dry regions and avoid sensitive areas thatmay cause distress. Skin was swabbed by moving aside hair or fur with
gloved hands to expose the skin. While applying moderate pressure, the skinwas swabbed 10 times in a forward and backward motion. The swab wasrotated and repeated in adjacent areas for a total of 40 strokes per swab.When the area was complete, the samples were returned to their initialplastic storage container and frozen at −20 °C until further use. All volun-teers and veterinary technicians were trained with a detailed protocol toensure sample collection consistency.
Sample Preparation. All DNA extractions, PCR protocols, and Illumina se-quencing were conducted according to a previously published protocol (63)to enable comparisons between the human and nonhuman samples. Inbrief, DNA was extracted using the PowerSoil-htp 96-well DNA Isolation Kit(MO BIO Laboratories) and stored at −20 °C until further use. The V3-V4 region of the 16S rRNA gene was amplified using the Pro-341Fi (5′-CCTACGGGNBGCASCAG-3′) and Pro-805Ri (5′-GACTACNVGGGTATCTAATCC-3′) primers (64). Each amplification was performed in triplicate to minimizepotential PCR bias from low biomass samples (65), and was conducted in aPCR hood that was UV-treated for 30 min after having undergone a treatmentwith UltraClean Lab Cleaner (MO BIO Laboratories) to remove DNA, RNA, DNase,and RNAses (66). No template controls, kit extraction controls, and sterile swabcontrols were included, in addition to including the same samples across allthree MiSeq runs to ensure lane-to-lane consistency (“run controls”). Addi-tional descriptions of control methods and results are included in SI Appendix(SI Appendix, Supplementary Information Text and Fig. S5) and Dataset S3.The Pro341F/Pro805R primer set was analyzed using TestPrime v1.0 againstthe SILVA database to assess potential sequence mismatches against bacterialand archaeal reference sequences (SI Appendix, Table S3) (67, 68).
Processing of Sequence Data. Raw DNA sequence reads were processed usingthe same open-source bioinformatics pipeline described previously (63) thatwasmanaged by Automation, eXtension, and Integration Of Microbial Ecology(AXIOME) v1.5 (69). PANDAseq v2.8 (70) generated paired-end sequences usingthe default parameters of a 0.9 quality threshold, a minimum sequence overlapof 10 bases, and a minimum read length of 300 bases. Quantitative InsightsInto Microbial Ecology (QIIME) v1.9.0 (71) was used to analyze sequence data,which underwent de novo clustering and chimera/singleton removal at both97% and 99% cluster identity using UPARSE (72). PyNAST v1.2.2 (73) was usedto align 16S rRNA gene sequences. Subsequently, RDP v8.1 (74) assigned pro-karyotic taxonomy using Greengenes database v13.8 (75). Samples were rar-efied to 1,654 sequences in the dataset that contained all samples (Dataset S1).Analyses such as α-diversity and PERMANOVA underwent 1,000 iterations ofrarefication to avoid underrepresenting diversity. Rarefication plots demon-strated that conducting multiple rarefications to determine diversity preventeda loss in diversity levels. Other mammalian skin microbiome studies have used asimilar level of sequences to analyze their data (9, 13, 76, 77).
Statistical Analyses. The majority of statistical analyses were conducted usingthe same programs and software version numbers as described previously(63). In brief, α-diversity of all samples was measured with the followingQIIME commands: multiple_rarefaction.py, alpha_diversity.py, collate_alpha.py, and compare_alpha_diversity.py. Subsequently, the 42 metadata cate-gories were compared using the Bonferroni correction to avoid false-positives due to testing a high number of hypotheses.
β-Diversity was visualized using ordinations generated with the Bray–Curtis distance metric. Ordinations were created in RStudio (78) with thephyloseq (v1.14.0) and ggplot2 (v2.1.0) packages. β-Diversity was measuredusing PERMANOVAwith the Adonis function from the vegan package (v2.4–0)in R. Using 1,000 permutations, the percent variation explained by eachmetadata category was calculated and visualized in a heatmap using ggplot2,vegan, Heatplus 2.16.0 (79), and RcolorBrewer v1.1.2 (80).
The functions from the prokaryotic cladeswere predicted using FAPROTAX(35). This conservative algorithm currently matches 80 functions, such asfermentation and methanogenesis, against 7,600 functional annotations of4,600 prokaryotic taxa.
Phylosymbiosis Analysis. The phylosymbiosis analysis of skin microbiota pro-files and host phylogeny was adapted from a previously described protocol(19). Downloaded COX1 sequences were aligned with Muscle v3.8.31 (81)and edited by removing gap positions and 5′/3′ end overhangs with Jalviewv2.9 (82). The final edited alignment was created using RaxML onlineBlackbox server v8.2 (83). All mammalian host trees were verified to be inconcordance with well-established mammalian phylogenies (84–88).
Microbiota dendrograms were constructed using the QIIME v1.9.0 jackknifed_beta_diversity.py command. Species were rarefied to the highest possible se-quence count that included all species within the specific taxonomic ranking.
Ross et al. PNAS | vol. 115 | no. 25 | E5793
MICRO
BIOLO
GY
PNASPL
US
Dow
nloa
ded
by g
uest
on
June
30,
202
0
This resulted in a rarefication of 1,900 sequences for all mammals, 9,100 forArtiodactyla, 25,700 for Carnivora, and 37,500 for Perissodactyla. Rar-efication was conducted 1,000 times and a consensus tree built to correctfor the low number of rarefied sequences. Each of the above mammalianclades had skin microbiota consensus dendrograms created at 97% and 99%OTU identity threshold using Bray–Curtis, weighted and unweighted UniFracdistance metrics.
Congruencies between host phylogenies and skin microbiota dendro-grams were quantified by calculating normalized Robinson–Foulds (89) andnormalized matching-cluster scores. These scores were calculated by usingthe ape R package (90) and a custom python script created by Brooks et al.(19), respectively. The significance of these scores was determined by con-structing 100,000 randomized trees with identical leaf nodes to the skinmicrobiota dendrograms and comparing each to the host phylogeny to
calculate the number of stochastic dendrograms with equivalent or betterRobinson–Foulds or matching cluster scores.
ACKNOWLEDGMENTS. We thank Rahgavi Poopalarajah, Mayar Zawawi, andElena Dybner for assistance with sample kit preparation and data entry;Katja Engel for support with Illumina library preparation; Jonathan Ross forassistance with Python; Charles Gray and his team from the African LionSafari; Dr. Diego Gomez from the University of Guelph; Andrea Dada, DawnMihailovic, and the veterinarian technicians from the Toronto Zoo; AmandaHawkins and Joe Growden from the Kitchener–Waterloo Humane Society;Heather Cray, Emilie Spasov, and all farmers and pet owners who partici-pated in the study; and Andrew Brooks for comments that improved thephylosymbiosis analysis. A.A.R. was supported by a Canada Graduate Schol-arship from the Canadian Institutes for Health Research. This research wassupported by a Discovery Grant to J.D.N. from the Natural Sciences andEngineering Research Council of Canada.
1. Woodhams DC, et al. (2014) Interacting symbionts and immunity in the amphibianskin mucosome predict disease risk and probiotic effectiveness. PLoS One 9:e96375.
2. Colston TJ, Jackson CR (2016) Microbiome evolution along divergent branches of thevertebrate tree of life: What is known and unknown. Mol Ecol 25:3776–3800.
3. Gomez D, Sunyer JO, Salinas I (2013) The mucosal immune system of fish: The evo-lution of tolerating commensals while fighting pathogens. Fish Shellfish Immunol 35:1729–1739.
4. Theis KR, Schmidt TM, Holekamp KE (2012) Evidence for a bacterial mechanism forgroup-specific social odors among hyenas. Sci Rep 2:615.
5. Shelley WB, Hurley HJ, Jr, Nichols AC (1953) Axillary odor; experimental study of therole of bacteria, apocrine sweat, and deodorants. AMA Arch Derm Syphilol 68:430–446.
6. Kuhn F, Natsch A (2009) Body odour of monozygotic human twins: A common patternof odorant carboxylic acids released by a bacterial aminoacylase from axilla secretionscontributing to an inherited body odour type. J R Soc Interface 6:377–392.
7. James AG, Austin CJ, Cox DS, Taylor D, Calvert R (2013) Microbiological and bio-chemical origins of human axillary odour. FEMS Microbiol Ecol 83:527–540.
8. Grice EA, Segre JA (2011) The skin microbiome. Nat Rev Microbiol 9:244–253.9. Council SE, et al. (2016) Diversity and evolution of the primate skin microbiome. Proc
Biol Sci 283:20152586.10. Avena CV, et al. (2016) Deconstructing the bat skin microbiome: Influences of the host
and the environment. Front Microbiol 7:1753.11. Apprill A, et al. (2014) Humpback whale populations share a core skin bacterial
community: Towards a health index for marine mammals? PLoS One 9:e90785.12. Apprill A, Mooney TA, Lyman E, Stimpert AK, Rappé MS (2011) Humpback whales
harbour a combination of specific and variable skin bacteria. Environ Microbiol Rep 3:223–232.
13. Rodrigues Hoffmann A, et al. (2014) The skin microbiome in healthy and allergic dogs.PLoS One 9:e83197.
14. Older CE, et al. (2017) The feline skin microbiota: The bacteria inhabiting the skin ofhealthy and allergic cats. PLoS One 12:e0178555.
15. Meason-Smith C, et al. (2017) Characterization of the cutaneous mycobiota in healthyand allergic cats using next generation sequencing. Vet Dermatol 28:71–e17.
16. Weese JS (2013) The canine and feline skin microbiome in health and disease. VetDermatol 24:137–145.e31.
17. Zinicola M, et al. (2015) Altered microbiomes in bovine digital dermatitis lesions, andthe gut as a pathogen reservoir. PLoS One 10:e0120504.
18. Cheng Y, et al. (2015) The Tasmanian devil microbiome-implications for conservationand management. Microbiome 3:76.
19. Brooks AW, Kohl KD, Brucker RM, van Opstal EJ, Bordenstein SR (2016) Phylo-symbiosis: Relationships and functional effects of microbial communities across hostevolutionary history. PLoS Biol 14:e2000225.
20. Moeller AH, et al. (2016) Cospeciation of gut microbiota with hominids. Science 353:380–382.
21. Fierer N, Hamady M, Lauber CL, Knight R (2008) The influence of sex, handedness,and washing on the diversity of hand surface bacteria. Proc Natl Acad Sci USA 105:17994–17999.
22. Grice EA, et al.; NISC Comparative Sequencing Program (2008) A diversity profile ofthe human skin microbiota. Genome Res 18:1043–1050.
23. Culp CE, Falkinham JO, Belden LK (2007) Identification of the natural bacterial mi-croflora on the skin of eastern newts, bullfrog tadpoles and redback salamanders.Herpetologica 63:66–71.
24. Price PB (1938) The bacteriology of normal skin: A new quantitative test applied to astudy of the bacterial flora and the disinfectant action of mechanical cleansing.J Infect Dis 63:301–318.
25. Lilly HA, Lowbury EJ (1978) Transient skin flora: Their removal by cleansing or disin-fection in relation to their mode of deposition. J Clin Pathol 31:919–922.
26. Glazko GV, Nei M (2003) Estimation of divergence times for major lineages of primatespecies. Mol Biol Evol 20:424–434.
27. Clemente JC, et al. (2015) The microbiome of uncontacted Amerindians. Sci Adv 1:e1500183.
28. Fitzpatrick BM, Allison AL (2014) Similarity and differentiation between bacteria as-sociated with skin of salamanders (Plethodon jordani) and free-living assemblages.FEMS Microbiol Ecol 88:482–494.
29. Kueneman JG, et al. (2014) The amphibian skin-associated microbiome across species,space and life history stages. Mol Ecol 23:1238–1250.
30. Bouslimani A, et al. (2015) Molecular cartography of the human skin surface in 3D.Proc Natl Acad Sci USA 112:E2120–E2129.
31. Oh J, et al.; NISC Comparative Sequencing Program (2014) Biogeography and in-dividuality shape function in the human skin metagenome. Nature 514:59–64.
32. Grice EA, et al.; NISC Comparative Sequencing Program (2009) Topographical andtemporal diversity of the human skin microbiome. Science 324:1190–1192.
33. Song SJ, et al. (2013) Cohabiting family members share microbiota with one anotherand with their dogs. eLife 2:e00458.
34. Dunn RR, Fierer N, Henley JB, Leff JW, Menninger HL (2013) Home life: Factorsstructuring the bacterial diversity found within and between homes. PLoS One 8:e64133.
35. Louca S, Parfrey LW, Doebeli M (2016) Decoupling function and taxonomy in theglobal ocean microbiome. Science 353:1272–1277.
36. Consortium HMP; Human Microbiome Project Consortium (2012) Structure, functionand diversity of the healthy human microbiome. Nature 486:207–214.
37. Lloyd-Price J, et al. (2017) Strains, functions and dynamics in the expanded HumanMicrobiome Project. Nature 550:61–66, and erratum (2017) 551:256.
38. Schwartz IL, Thaysen JH, Dole VP (1953) Urea excretion in human sweat as a tracer formovement of water within the secreting gland. J Exp Med 97:429–437.
39. Deano P, Robinson JW (1986) Direct determination of manganese in perspiration andurine using atomic absorption spectroscopy. Spectrosc Lett 19:11–19.
40. Cohn JR, Emmett EA (1978) The excretion of trace metals in human sweat. Ann ClinLab Sci 8:270–275.
41. Zhang LM, et al. (2008) Microbial DNA extraction and analyses of soil iron-manganesenodules. Soil Biol Biochem 40:1364–1369.
42. Luo J-D, Wang Y-Y, Fu W-L, Wu J, Chen AF (2004) Gene therapy of endothelial nitricoxide synthase and manganese superoxide dismutase restores delayed wound heal-ing in type 1 diabetic mice. Circulation 110:2484–2493.
43. Lener T, et al. (2006) Expression profiling of aging in the human skin. Exp Gerontol 41:387–397.
44. Gill SR, et al. (2006) Metagenomic analysis of the human distal gut microbiome.Science 312:1355–1359.
45. LeBlanc JG, et al. (2013) Bacteria as vitamin suppliers to their host: A gut microbiotaperspective. Curr Opin Biotechnol 24:160–168.
46. Foster JA, McVey Neufeld KA (2013) Gut-brain axis: How the microbiome influencesanxiety and depression. Trends Neurosci 36:305–312.
47. Bogdanowicz D, Giaro K, Wróbel B (2012) TreeCmp: Comparison of trees in poly-nomial time. Evol Bioinf Online 8:475–487.
48. Novakova E, et al. (2017) Mosquito microbiome dynamics, a background for preva-lence and seasonality of West Nile virus. Front Microbiol 8:526.
49. Ley RE, et al. (2008) Evolution of mammals and their gut microbes. Science 320:1647–1651.
50. Antwis RE, et al. (2014) Ex situ diet influences the bacterial community associatedwith the skin of red-eyed tree frogs (Agalychnis callidryas). PLoS One 9:e85563.
51. Ochman H, et al. (2010) Evolutionary relationships of wild hominids recapitulated bygut microbial communities. PLoS Biol 8:e1000546.
52. Gaci N, Borrel G, Tottey W, O’Toole PW, Brugère JF (2014) Archaea and the humangut: New beginning of an old story. World J Gastroenterol 20:16062–16078.
53. Hassanshahian M, Mohamadian J (2011) Isolation and characterization of Halobacteriumsalinarum from saline lakes in Iran. Jundishapur J Microbiol 4:59–66.
54. Probst AJ, Auerbach AK, Moissl-Eichinger C (2013) Archaea on human skin. PLoS One8:e65388.
55. Moissl-Eichinger C, et al. (2017) Human age and skin physiology shape diversity andabundance of Archaea on skin. Sci Rep 7:4039.
56. Raymann K, Moeller AH, Goodman AL, Ochman H (2017) Unexplored archaeal di-versity in the great ape gut microbiome. MSphere 2:e00026-17.
57. Kong HH (2016) Details matter: Designing skin microbiome studies. J Invest Dermatol136:900–902.
58. Zeeuwen PLJM, et al. (2017) Reply to Meisel et al. J Invest Dermatol 137:961–962.59. Meisel JS, et al. (2016) Skin microbiome surveys are strongly influenced by experi-
mental design. J Invest Dermatol 136:947–956.60. Conlan S, Kong HH, Segre JA (2012) Species-level analysis of DNA sequence data from
the NIH Human Microbiome Project. PLoS One 7:e47075.61. Castelino M, et al. (2017) Optimisation of methods for bacterial skin microbiome in-
vestigation: Primer selection and comparison of the 454 versus MiSeq platform. BMCMicrobiol 17:23.
E5794 | www.pnas.org/cgi/doi/10.1073/pnas.1801302115 Ross et al.
Dow
nloa
ded
by g
uest
on
June
30,
202
0
62. Metcalf JL, et al. (2016) Microbial community assembly and metabolic function duringmammalian corpse decomposition. Science 351:158–162.
63. Ross AA, Doxey AC, Neufeld JD (2017) The skin microbiome of cohabiting couples.mSystems 2:e00043-17.
64. Takahashi S, Tomita J, Nishioka K, Hisada T, Nishijima M (2014) Development of aprokaryotic universal primer for simultaneous analysis of bacteria and archaea usingnext-generation sequencing. PLoS One 9:e105592.
65. Kennedy K, Hall MW, Lynch MDJ, Moreno-Hagelsieb G, Neufeld JD (2014) Evaluatingbias of illumina-based bacterial 16S rRNA gene profiles. Appl Environ Microbiol 80:5717–5722.
66. Galan M, et al. (2016) 16S rRNA amplicon sequencing for epidemiological surveys ofbacteria in wildlife. mSystems 1:e00032-16.
67. Pruesse E, et al. (2007) SILVA: A comprehensive online resource for quality checkedand aligned ribosomal RNA sequence data compatible with ARB. Nucleic Acids Res 35:7188–7196.
68. Klindworth A, et al. (2013) Evaluation of general 16S ribosomal RNA gene PCR pri-mers for classical and next-generation sequencing-based diversity studies. NucleicAcids Res 41:e1.
69. Lynch MDj, Masella AP, Hall MW, Bartram AK, Neufeld JD (2013) AXIOME: Auto-mated exploration of microbial diversity. Gigascience 2:3.
70. Masella AP, Bartram AK, Truszkowski JM, Brown DG, Neufeld JD (2012) PANDAseq:Paired-end assembler for Illumina sequences. BMC Bioinformatics 13:31.
71. Caporaso JG, et al. (2010) QIIME allows analysis of high-throughput community se-quencing data. Nat Methods 7:335–336.
72. Edgar RC (2013) UPARSE: Highly accurate OTU sequences from microbial ampliconreads. Nat Methods 10:996–998.
73. Caporaso JG, et al. (2010) PyNAST: A flexible tool for aligning sequences to a templatealignment. Bioinformatics 26:266–267.
74. Wang Q, Garrity GM, Tiedje JM, Cole JR (2007) Naive Bayesian classifier for rapidassignment of rRNA sequences into the new bacterial taxonomy. Appl EnvironMicrobiol 73:5261–5267.
75. McDonald D, et al. (2012) An improved Greengenes taxonomy with explicit ranks forecological and evolutionary analyses of bacteria and archaea. ISME J 6:610–618.
76. Hulcr J, et al. (2012) A jungle in there: Bacteria in belly buttons are highly diverse, butpredictable. PLoS One 7:e47712.
77. Ying S, et al. (2015) The influence of age and gender on skin-associated microbialcommunities in urban and rural human populations. PLoS One 10:e0141842.
78. RStudio Team (2016) RStudio: Integrated Development for R (RStudio, Inc., Boston),Version 0.99.896.
79. Ploner A (2015) Heatplus: Heatmaps with row and/or column covariates and col-ored clusters. R package version 2.26.0. Available at https://github.com/alexploner/Heatplus. Accessed March 17, 2017.
80. Harrower M, Brewer CA (2011) ColorBrewer.org: An online tool for selecting colourschemes for maps. Cartogr J 40:27–37.
81. Edgar RC (2004) MUSCLE: Multiple sequence alignment with high accuracy and highthroughput. Nucleic Acids Res 32:1792–1797.
82. Waterhouse AM, Procter JB, Martin DMA, Clamp M, Barton GJ (2009) Jalview version2–A multiple sequence alignment editor and analysis workbench. Bioinformatics 25:1189–1191.
83. Stamatakis A (2006) RAxML-VI-HPC: Maximum likelihood-based phylogenetic analy-ses with thousands of taxa and mixed models. Bioinformatics 22:2688–2690.
84. Delisle I, Strobeck C (2005) A phylogeny of the Caniformia (order Carnivora) based on12 complete protein-coding mitochondrial genes. Mol Phylogenet Evol 37:192–201.
85. Agnarsson I, Kuntner M, May-Collado LJ (2010) Dogs, cats, and kin: A molecularspecies-level phylogeny of Carnivora. Mol Phylogenet Evol 54:726–745.
86. Li WH, Gouy M, Sharp PM, O’hUigin C, Yang YW (1990) Molecular phylogeny ofRodentia, Lagomorpha, Primates, Artiodactyla, and Carnivora and molecular clocks.Proc Natl Acad Sci USA 87:6703–6707.
87. Hassanin A, Douzery EJP (1999) The tribal radiation of the family Bovidae (Artiodactyla)and the evolution of the mitochondrial cytochrome b gene. Mol Phylogenet Evol 13:227–243.
88. Perelman P, et al. (2011) A molecular phylogeny of living primates. PLoS Genet 7:e1001342.
89. Robinson DF, Foulds LR (1981) Comparison of phylogenetic trees. Math Biosci 53:131–147.
90. Paradis E, Claude J, Strimmer K (2004) APE: Analyses of phylogenetics and evolution inR language. Bioinformatics 20:289–290.
Ross et al. PNAS | vol. 115 | no. 25 | E5795
MICRO
BIOLO
GY
PNASPL
US
Dow
nloa
ded
by g
uest
on
June
30,
202
0
top related